As covered in the last blog post, when we first built the individual ID workflow, we built it around the most-common use case we had encountered – where an organisation will perform a discrete survey of the population of a particular species using spatially-explicit capture recapture. In particular, for more prevalent species like spotted hyena, there would be thousands of images of hundreds of different individuals. Moreover, these images would be processed by students or interns that would not know any of the individuals by heart. As such, the process of individual ID was simply a giant pattern-matching exercise subject to combinatorial growth where the name of the game was to simply reduce the number of image pairs that would need to be reviewed.
Considering this, the individual ID workflow was built around the idea of performing ID independently on a survey first. Thereafter, once it was reduced from a few thousand images into a few hundred individuals, it could then be identified against another ID-complete survey with its few hundred individuals. This served both the purpose of keeping the combinatorial growth in check, and allowing users to do their SECR analysis prior to this step.
However, this would obviously be frustrating and inefficient for organisations performing long-term monitoring of a protected area. Firstly, the long-term annotators who were able to identify an individual by sight would not be able to do anything about it during the workflow, and instead be forced to follow the strict pattern-matching exercise. This was dealt in part one of this update – by allowing users to push an animal to a known individual at the very start of the ID process. Secondly, the old approach would not take advantage of the long-term library of individuals – both in terms of the additional context provided by years of sightings/images, and the greater matching opportunities for the AI algorithm.
Therefore we have added the concept of areas to the platform – where a survey can optionally be associated with a particular protected area or wider unfenced region where individuals are known to roam. This firstly facilitates search and organisation of these surveys, but also allows organisations to build up libraries of known individuals in discrete areas or parks. Therefore, when a new survey is added to the platform, it can now either undergo individual ID independently or directly against the library of individuals in the area based on your selection on the individual ID launch menu as shown in Figure 1.
Figure 1: Users now have the choice to ID a new survey independently or against the library of individuals for its area.
In the case of the latter (identifying against your library), you will still need to initially perform the first stage of the individual ID workflow where you need to associate the individuals within a cluster first – providing as many view angles as possible for the AI algorithm. Thereafter, in the second stage TrapTagger can take advantage of a well-established library and provide better AI suggestions, thus allowing you to more-rapidly consolidate your new data into your library.
Note: By default, pre-existing surveys have been left with no defined area, which you can modify by clicking the edit-survey button, switching to the area tab and making the necessary changes – ie. creating a new area, or selecting a pre-existing one. For surveys where organisations have already built up a library of individuals across multiple surveys, those surveys have been automatically grouped together under an auto-generated area. You can edit this name by following the same steps.
Individual ID Tools
One has always been able to add notes and informational tags (male/female, adult/juvenile etc.) to individuals to aid in the identification process. However, these weren’t displayed during the giant pattern-matching workflow. As part of this long-term-library update, we have made this information easily-referenceable as collapsible side-panels during the main individual ID workflow (part 2 where you combine cluster individuals into actual individuals). As depicted in Figure 2, we have also extended this individual information to include a number of different things: the areas in which an individual has been seen, the first and last time it was seen, and highlighted features drawn onto the individual. Together, these aid collaborative ID between multiple annotators by allowing people to make note of the various features they look for to identify a specific individual.
Figure 2: The new collapsible individual-information side panels available during the ID workflow.
Moreover, we have improved the various tools around the individual ID process in a number of ways:
- There are now flank filters during the individual ID workflow that filter the two currently-displayed individuals to only show a particular flank – which can be quite useful when working with individuals with large numbers of images.
- The sightings/bounding boxes that you have marked as unidentifiable can now be found on the individuals library where you can now restore them as needed.
- You can mark a particular sighting/bounding box associated with an individual as unidentifiable from that individual’s page in the library.
- You can also edit the flank of a sighting/bounding box from the associated individual’s page by right-clicking on it.
- The primary images for an individual can be edited from its library page.
- The features drawn on these primary images can be edited there too.
- More individual-level analysis tools have been added to their pages – such as activity patterns and time-based sightings.
- Images of individuals will now be permanently de-archived to facilitate work across your library.
Miscellaneous
There are also a number of small tweaks and features that we have added that are tangentially (or not) related to this update:
- Survey areas can be used for filtering on the surveys page.
- Surveys now have a start and stop date (extracted from their first and last files). This can be used for easy filtering of surveys by year – which can be quite useful in combination with the area filter.
- Survey areas have been added as a filter option across all analysis tools.